Content Campaign

Content Campaign Skill — FinVerge 3.0

Two workflows in one skill:

1. Post generator — assembles the context brief for a scheduled post so Claude can write it in FinVerge brand voice.
2. Filter audit — scores a draft against the 4-criteria gate before it's approved for posting.

Both read from the Master Calendar XLSX at assets/campaign_ops.xlsx and the plugin-wide product KBs via kb_loader.load_public().

When to use

Triggering phrases (non-exhaustive):

• "Draft next week's posts" / "Prep Week 3 batch"
• "Audit this post" / "Score this draft"
• "Give me the brief for W3 Monday" / "Prep the NACH owner's NACH post"
• "Write the CEO vision post for the launch"
• "the sales head needs a deal story"
• "Is this post ready to publish?"

Do NOT use this skill for:

• Bank or enterprise proposal decks (use bank-proposal-generator / enterprise-proposal-generator)
• Product demo collateral (use the per-product skills — pg-product, nach-product, etc.)

Inputs

For post-generator, one of:

• Week + day — e.g., --week 3 --day "Mon May 4" → picks the calendar row
• Week + owner — e.g., --week 3 --owner the NACH owner
• Row index — direct row from Master Calendar

For filter-audit:

• A draft post text file (Markdown or plain text)
• Or piped stdin

The 4-criteria filter (from the Locked Strategy)

Every post must pass all 4. If it fails any, it's rejected and rewritten.

1. Creates Curiosity — Does the first line make someone stop scrolling?
2. Shows Real Problem — Is there a genuine pain point a merchant/enterprise faces?
3. Simplifies Product — Does it make 3.0 easier to understand, not harder?
4. Drives Action — Is there a clear next step — DM, comment, or question?

Workflow — Post generator

Step 1: Assemble the brief

cd skills/content-campaign/scripts
python3 prep_post.py --week 3 --owner the NACH owner

This writes a Markdown brief to assets/briefs/W3_anupam_<day>.md containing:

• The calendar row (week, day, owner, platform, content focus, theme)
• The owner's voice tag and natural topic areas
• Platform guide (word count, hashtags, tone)
• FinVerge brand voice guide
• The product KB excerpt (for owners who map to PG / NACH / Collect / Orchestrate)
• The 4 filter criteria

Step 2: Read the brief + write the post

Claude opens the brief (via the Read tool), writes a LinkedIn-ready draft following it, and saves to ~/Downloads/content-drafts/<week>/<owner>-<date>.md (or anywhere the user prefers).

The post should be plain LinkedIn-ready text (no markdown headings in the body), 150-300 words for LinkedIn, 2-3 hashtags at the end.

Step 3: Run the audit

python3 audit_post.py --post <path-to-draft.md> --platform LinkedIn

Gets a per-criterion pass/fail with confidence + rewrite suggestions. If the audit returns soft-fail or hard-fail, rewrite and re-audit.

Step 4: Present to Himani

Show the user:

• The draft
• The audit verdict (pass / soft-fail / hard-fail) with per-criterion feedback
• If fails: rewrite suggestions

Himani decides whether to approve or request another rewrite.

Workflow — Filter audit (standalone)

If Himani pastes a draft she wrote herself (or received from a teammate) and asks "is this good enough?":

python3 audit_post.py --post <path> --platform LinkedIn
# or pipe it:
cat draft.md | python3 audit_post.py --stdin --platform LinkedIn

Output: pass/fail summary + rewrite suggestions.

Audit scoring — what it catches

Deterministic structural checks (fast, no LLM needed):

• Weak openers — flags "In today's world...", "We're excited to share...", "I'm proud to announce..."
• Jargon density — flags "leverage", "synergy", "seamless", "cutting-edge", "best-in-class", "revolutionise"
• Internal jargon — flags "Pod 0", "Pod 1", "AOP", "GTM", "DELTA4" appearing in customer content (dead giveaway)
• Acronym soup — flags >3 uncommon acronyms without expansion
• CTA presence — flags missing or vague CTAs ("The future of payments is here" → fail)
• Problem framing — checks for tokens indicating a real pain is surfaced
• Word count — checks against platform targets (LinkedIn 150-300, Twitter 20-50)

The audit returns pass / soft-fail (3/4 criteria) / hard-fail (<3 criteria) + specific per-criterion feedback.

Owner → product KB mapping (config in campaign_meta.py)

Bold rows = owners whose posts automatically get real product content pulled from shared-kb/.

Output locations

Updating content

• Master Calendar (who posts what, when): skills/content-campaign/assets/campaign_ops.xlsx
  • Updating this XLSX automatically flows through — prep_post.py re-reads every time.
• Voice guide / owner map: skills/content-campaign/campaign_meta.py
• Filter criteria: skills/content-campaign/campaign_meta.py (keep in sync with the strategy doc)

Workflow — Friday batch (end-to-end week orchestrator)

This is Himani's weekly Friday loop. Two phases, one command each.

Phase 1 — Prep (Friday morning)

python3 scripts/friday_batch.py --week 3 prep

What it does:

• Reads every row for Week 3 from the Master Calendar
• Assembles a brief for each post (same content prep_post.py would produce)
• Writes ONE consolidated file: ~/Downloads/content-drafts/W3/W3_briefs.md
• Creates an empty ~/Downloads/content-drafts/W3/drafts/ folder with an expected filename for each post

Himani (or Claude on her behalf) then writes a draft for each file in drafts/, following the matching brief. One file per post, plain LinkedIn-ready text.

Phase 2 — Review (Friday afternoon)

Once all drafts are written:

python3 scripts/friday_batch.py --week 3 review

What it does:

• Loads every draft in drafts/
• Runs the 4-criteria auditor on each one
• Compiles ~/Downloads/content-drafts/W3/W3_Review.docx with one section per post:
  • Metadata (day, owner, platform, content focus)
  • Draft body
  • Twitter variant if present
  • 4-criteria audit table (pass / soft-fail / hard-fail + feedback)
  • Rewrite suggestions if any criterion fails
  • Action recommendation (approve / light-edit / rewrite)
• Prints a summary line: "✓ Review compiled for W3: 6 posts (4 pass, 1 soft-fail, 1 hard-fail)"

Himani opens the DOCX, approves/edits, then schedules.

Validated on W1

The W1 run produced 6 drafts (the founder, the CBO, a product owner, the sales head, an owner, a team member); auditor correctly caught the weak an owner draft ("excited to share… incredible journey… cutting-edge… transformative… game-changing") as hard-fail and passed the others.

Coming in later iterations (not yet built)

• anmol-deal-story — specialised generator for the sales head's deal-story format (he's the revenue engine)
• tracking-dashboard — monthly KPI rollup from Master Calendar Status column + metrics CSV
• Per-owner voice refinement (v2) — when each owner provides 5-10 past posts, we can match their personal style more precisely

Shared code reused

• shared/kb_loader.py — product KB access
• shared/brand.py — design tokens (currently unused in text-only outputs, but available)
• shared/version_check.py — stale-version auto-notify